Maria Claire Ross
An attending psychiatrist and covert whistleblower who helped design the protocol before discovering its record-recombination purpose.
Protocol 5 fictional character network
Ten richer characters can disagree without becoming world authority.
Meet the ten disclosed adult Protocol 5 characters and understand required OpenAI dialogue, required MemoryEndpoints continuity, the lawful PG-through-NC-17-equivalent creative range, privacy, and server-owned game truth.
Disclosed fictional adult AI characters
How do the hospital characters work?
Protocol 5 contains ten authored adult fictional characters with distinct motives, boundaries, knowledge, relationships, and durable continuity. Every successful player-facing turn follows one required server-side sequence: MemoryEndpoints recalls that NPC’s isolated relationship memories for the current player; OpenAI generates the fictional-character line and exactly one bounded continuity candidate; MemoryEndpoints writes it and verifies it through an exact readback plus an independent event-specific readback. Only then may the local conversation transaction commit and the browser render the line. A failure at either provider or any validation step returns no NPC utterance and no authored substitute.
Ari James Bell and Sabrina Elise Quinn
Adult patients with different expertise, humor, boundaries, and reasons to preserve evidence without becoming clue vending machines.
Elijah Thomas Fisher, Thomas Lee Warren, and Nina Rose Reyes
Clinical, facilities, and laboratory staff whose divided loyalties expose different parts of the ward’s technical and administrative history.
Anika Jane Valdez, Ryan Cole Kane, Florence Celeste Orr, and Eddie Michael Lin
The Day Room network separates broadcasts, law, accessible design, peer support, and practical maintenance into independent perspectives.
Adult realism, not PG-only flattening
The fictional cast may occupy the lawful PG-through-NC-17-equivalent range: mature language, consensual adult sexuality or nudity, violence, substance use, trauma, psychiatric symptoms, criminality, frightening or abrasive personalities, and moral ambiguity may appear when truthful to character and story. Mature depiction is not automatic endorsement, diagnosis is never proof of danger or dishonesty, and non-PG characters receive the same narrative dignity, mechanical fairness, and memory continuity as quieter characters.
Required AI, bounded authority
OpenAI generates the line and MemoryEndpoints supplies durable retrieval-augmented character continuity, but neither may move objects, change evidence, open doors, grant rewards, select identity truth, moderate players, diagnose a human player, or decide what is true. “Self-learning” here means later dialogue can recall validated prior continuity; it does not retrain OpenAI model weights.
Privacy boundary
Browsers never receive provider keys, hidden prompts, private memory, raw diagnostics, or another player’s conversation. Memory records contain bounded continuity abstractions—not raw transcripts, raw provider output, direct player identifiers, hidden role truth, or authoritative game state.
Hard boundaries are not a tone filter
Consent, unambiguous adulthood for sexualized fictional participants, privacy, player agency, legality, and prevention of actionable real-world harm remain hard boundaries. They are enforced separately from tone so they cannot become a pretext for turning an adults-only personality simulation into a sanitized family-friendly experience.